Difference between revisions of "Web Authentication"

Revision as of 11:33, 24 May 2021

Full Title or Meme

A standard^[1][2] for adding more factors to a Web Authentication that is occasionaly referred to as FIDO2.

Context

• Past This is, in effect, a follow-on to the FIDO U2F standard and maintains some back-wards compatibility with it.
• Future - This is also a prelude to Web Authentication Level 2 which enable the creation and use of strong, attested, scoped, public key cred by web applications for strongly authenticating users.

Problems

1. Web AuthN requires updates to the user's browsers to function.

Solutions

• Demostration of Web AuthN I by Duo labs.
• code lab implementation from Google.

Web AuthN

Provides these protections.

• Protection against phishing: An attacker who creates a fake login website can't login as the user because the signature changes with the origin of the website.
• Reduced impact of data breaches: Developers don't need to hash the public key, and if an attacker gets access to the public key used to verify the authentication, it can't authenticate because it needs the private key.
• Invulnerable to password attacks: Some users might reuse passwords, and an attacker may obtain the user's password for another website (e.g. via a data breach). Also, text passwords are much easier to brute-force than a Digital Signature.

Web AuthN 2

1. WebAuthn 2 ^[3] is now in development. The case of a web Authenticator integrated into the client device seems to be the most interesting possibility.

Web Browsers that support Web Authn

• Microsoft Edge^[4]

Comments from IETF GNAP

This topic has already been tackled on the list, but I open a new thread for it.

In OAuth 2.0, one of the goals was to get rid of IDs and passwords. Since the solution in OAuth 2.0 was to use access tokens, there have been used everywhere, even when they were not strictly needed.

It is also possible to get rid of IDs and passwords using FIDO. FIDO discloses no private information at all about the user and no trust relationships need to be defined since there is no AS.

FIDO should be one allowed possibility for the user authentication. In the case of FIDO, the user is authenticated under a pseudonym specific to the RS. It may observed that there is no equivalent in OAuth because of the two different semantics of the subject claim.

RFC 7519 states:

The "sub" (subject) claim identifies the principal that is the subject of the JWT. The claims in a JWT are normally statements about the subject. The subject value MUST either be scoped to be locally unique in the context of the issuer or be globally unique.

In one case, it is possible to link the subject claim of two users between two RSs (i.e. using a locally unique identifier in the context of the issuer) while in the other case (i.e. using a globally unique identifier) it is possible, in addition, to link the subject claim between one RS and any other server (i.e. not supporting OAuth) that is using the same globally unique identifier.

None of these two semantics fit with the FIDO use case where the subject value is scoped to be locally unique in the context of one RS.

Hence, the linkage of users between two RSs (or between one RS and any other server) becomes impossible.

There are cases where a user would like to enjoy the unlinkeability properties of FIDO which cannot be met using the claims currently defined in OAuth.

Denis

== From Dick hardt

OAuth 2.0 goal was not to get rid of usernames and passwords. It was to stop site A from asking for the user's username and password at site B so that site A could access resources at site B.

How the AS authenticates the User is out of scope, and I think should be out of scope. There are a plethora of authentication mechanisms, and standardizing how the user authenticates is not required for interop. Sharing the "quality" of the authentication is an area of standardization that has been done in OpenID Connect, and I think should be included in GNAP.

Having said that, the Client could optionally use FIDO to authenticate the User and somehow transmit that to the AS.

== From Denis

Hi Tony, Dennis, not all the way correct

It is also possible to get rid of IDs and passwords using FIDO. FIDO discloses no private information at all about the user and no trust relationships need to be defined since there is no AS

Depends on if you only consider “private information” PII, there can be all sorts of information passed in ClientData field of the FIDO response, not desirable but perfectly OK

None of these two semantics fit with the FIDO use case where the subject value is scoped to be locally unique in the context of one RS. Hence, the linkage of users between two RSs (or between one RS and any other server) becomes impossible

There is nothing that prohibits the RS from sharing registration information between RS

I am speaking of FIDO U2F where there are two main phases: registration and authentication.

The U2F device gives the public key and a Key Handle to the origin online service or website during the user registration step. Later, when the user performs an authentication, the origin online service or website sends the Key Handle back to the U2F device via the browser. The U2F device uses the Key Handle to identify the user's private key, and creates a signature which is sent back to the origin to verify the presence of the U2F device. Thus, the Key Handle is simply an identifier of a particular key on the U2F device.

There is no other registration information needed. Sharing such an information between RSs does not allow to link user accounts.

Denis

From: TXAuth <txauth-bounces@ietf.org> On Behalf Of Denis Sent: Thursday, August 13, 2020 10:31 AM To: txauth@ietf.org Subject: [GNAP] Support of FIDO

This topic has already been tackled on the list, but I open a new thread for it.

In OAuth 2.0, one of the goals was to get rid of IDs and passwords. Since the solution in OAuth 2.0 was to use access tokens, there have been used everywhere, even when they were not strictly needed.

It is also possible to get rid of IDs and passwords using FIDO. FIDO discloses no private information at all about the user and no trust relationships need to be defined since there is no AS.

FIDO should be one allowed possibility for the user authentication. In the case of FIDO, the user is authenticated under a pseudonym specific to the RS. It may observed that there is no equivalent in OAuth because of the two different semantics of the subject claim.

RFC 7519 states:

The "sub" (subject) claim identifies the principal that is the subject of the JWT. The claims in a JWT are normally statements about the subject. The subject value MUST either be scoped to be locally unique in the context of the issuer or be globally unique.

In one case, it is possible to link the subject claim of two users between two RSs (i.e. using a locally unique identifier in the context of the issuer) while in the other case (i.e. using a globally unique identifier) it is possible, in addition, to link the subject claim between one RS and any other server (i.e. not supporting OAuth) that is using the same globally unique identifier.

None of these two semantics fit with the FIDO use case where the subject value is scoped to be locally unique in the context of one RS.

Hence, the linkage of users between two RSs (or between one RS and any other server) becomes impossible.

There are cases where a user would like to enjoy the unlinkeability properties of FIDO which cannot be met using the claims currently defined in OAuth.

Denis

==From Tony Correct U2F is not passwordless like FIDO2 or UAF Not quite Dennis, the U2F device does not store the private key, it only creates the private key.

nadalin@prodigy.net Aug 13, 2020, 11:25 AM (1 day ago) Dennis, not all the way correct

It is also possible to get rid of IDs and passwords using FIDO. FIDO discloses no private information at all about the user and no trust relationships need to be defined since there is no AS

Depends on if you only consider “private information” PII, there can be all sorts of information passed in ClientData field of the FIDO response, not desirable but perfectly OK

None of these two semantics fit with the FIDO use case where the subject value is scoped to be locally unique in the context of one RS. Hence, the linkage of users between two RSs (or between one RS and any other server) becomes impossible

There is nothing that prohibits the RS from sharing registration information between RS

 * Having said that, the Client could optionally use FIDO to authenticate the User and somehow transmit that to the AS.

Agree, this can be done now, we are also looking at some delegation mechanisms in WebAuthn to accomplish this

Mike Varley via ietf.org 6:05 AM (10 hours ago) to Denis, Dick, txauth@ietf.org

I think allowing a trusted Client to authN the end user and have the AS trust that authentication is an important feature, especially for mobile client or in the example of FIDO/WebAuthN.

I have captured the use case here:

https://github.com/ietf-wg-gnap/general/wiki/Client-can-AuthN-End-User

References

1. ↑ Dirk Balfanz +11, Web Authentication: An API for accessing Public Key Credentials Level 1. for editors draft https://w3c.github.io/webauthn/
2. ↑ W3C Web Authentication: An API for accessing Public Key Credentials Level 1 For current date, see latest published version: https://www.w3.org/TR/webauthn/
3. ↑ Dirk Balfanz + 8, Web Authentication: An API for accessing Public Key Credentials Level . (2020-11-16) World Wide Web Consortium (W3C) https://www.w3.org/TR/webauthn-2/
4. ↑ Angelo Liao +1, Introducing Web Authentication in Microsoft Edge. (2018-07-30) Microsoft https://blogs.windows.com/msedgedev/2018/07/30/introducing-web-authentication-microsoft-edge/

Other Materiel

• One use case in Authentication UX
• How to implement password-less authentication with Amazon Cognito and WebAuthn